A novel method for untrained detection of compound fault in rolling bearing via fast Fourier Transform-Transformer model

Rotating machinery relies heavily on rolling bearings, which are vulnerable to compound faults involving multiple interacting failure modes. Traditional diagnostic methods often inadequately decouple these superimposed vibration patterns and lack adaptability to untrained fault categories. This study proposes a novel compound fault diagnosis model based on the Fast Fourier Transform-Transformer (FFT-Transformer) architecture, utilizing attention mechanisms to extract fault features from vibration signals. The model first applies FFT to isolate fault-related frequency bands, eliminating noise interference. A multi-head attention mechanism then deciphers temporal dependencies in vibration signals, enabling precise identification of coexisting faults without prior knowledge of compound patterns. Crucially, the compound fault discrimination terms dynamically classify untrained fault types by evaluating classifier confidence levels, circumventing the need for exhaustive training data. Experimental results demonstrate that the proposed method effectively identifies fault conditions absent from the training data, significantly improving diagnostic performance and model reliability. This approach represents a notable advancement in fault diagnosis for rotating machinery, offering a robust solution to the challenges of compound fault identification with minimal data requirements.